Please find the link and protocol for the following method:

"CORE SAMPLE" PCR: A method to re-PCR unique bands from products of mixed size
By: Ed Rybicki

http://www.mcb.uct.ac.za/manual/corepcr.htm

INTRODUCTION
The products of a PCR reaction - especially when this is done on eukaryotic genomic DNA, and when using degenerate primers - often contain a mixture of discrete-sized bands, one of which is the "right" one, while the others represent products of "non-specific" priming. It can be a problem to obtain the correct band in any state approaching purity while maintaining yield, and attempting to purify the band by cloning all the reaction products and then probing the library for the correct DNA can be extraordinarily tedious.
I have applied a simple "core sampling" procedure - involving "coring" an agarose sample out of a gel, and using it as template for another round of PCR - to get around this problem, and obtain unique bands from initially messy backgrounds. Of course, having a visible band of the size expected does help; however, the technique may be used on faith on "right-sized" invisible bands if need be.

PROTOCOL
1. Run products of a PCR amplification on 1-2% TBE agarose gel, as two or more replicate lanes.
2. Cut off 1 lane - flanked by marker DNA if desired, and notched to allow re-orientation with remainder of gel - and stain in preferred ethidium bromide concentration (I use 50 ng/ml for 10 min).
3. View excised stained piece on 254nm UV box for maximum senssitivity; notch or stab correct band(s) in sample lane.
4. Prepare "core samplers": using gloves and sterile scissors and cut off about 5mm from the tip of as many sterile yellow pipette tips (we use Gilson tips) as you will need for samples.
5. Align stained marked segment with remainder of gel. Use "core samplers" to stab out one or more cores of agarose from the centre of bands of interest, using stabbed/notched gel lane as reference: a standard gel should give about 10ul per core.
6. Stain remainder of gel, view and photograph at 254nm to ensure correct regions were sampled.
NOTE: IT IS POSSIBLE TO QUICKLY CORE A STAINED GEL DIRECTLY ON A 305 OR EVEN A 254 NM UV BOX; HOWEVER, MORE THAN A FEW SECONDS OF EXPOSURE RESULTS IN CROSS-LINKING AND NO AMPLIFICATION
7. Use core samples as substrate in PCR reactions: I make up 40ul/reaction of reaction mix, and allow 10ul per core. Simply add core to mix, vortex a little, spin down, cover with mineral oil. PCR according to taste (not inhibited by presence of a little bromophenol blue or of 50ng/ml ethidium bromide).
8. At end of PCR: if you allow the tubes to cool down the reaction mix will set: 2%-odd agarose diluted 1/5 sets quite well! This is no problem for gel running as you then end the PCR on a 10 min 72oC cycle, and load the sample into wells of a gel BEFORE submerging the gel: sample will set in the wells and not float out.
9. If you wish to extract DNA, end at 72oC and add 50ul pre-warmed phenol / 8-OH-quinoline and vortex, add 100ul chloroform / isoamyl alcohol (24:1), vortex, spin: agarose should be in the phenol/CHCl3 phase. ALTERNATIVELY: take off mineral oil using 50ul CHCl3, take out plug of solidified sample and wash in TE, then put into 0.5ml Eppendorf-type vial with some siliconised glass wool at bottom, and a small needle hole. Put little Eppi in big Eppi without a lid, and spin 6000 rpm 10 min (a la Heery et al., 1990; TIG 6(6):173). Collect filtrate, clean up by phenol/CHCl3 and isopropanol/ammonium acetate ppte (1 vol IP, 0.2 vol of 10M ammonium acetate).

COMMENTS
I have successfully re-amplified a unique 500bp band from a background of many bands up to 1.5kb from a cDNA PCR of cauliflower mosaic virus 35S RNA in total turnip RNA extract, and a 150bp band from a background of bands going up to 3kb from an amplification of Arabidopsis total genomic DNA using thoroughly degenerate primers - in the latter case, to a point where it could be sequenced directly (using same primers) after a subsequent amplification after purification from a gel plug as above.
The method has advantages over a previously-described toothpicking procedure in that a core sample is generally of defined volume, may be stored indefinitely, and provides material for multiple re-amplifications.